HD RP vs Built-In Pipeline Performance comparison

Already sent here

You can download latest one here :
https://drive.google.com/file/d/1XVz-SmNe3Bxc3pEGEBKeeYM87Dpihk0g/view?usp=sharing
https://drive.google.com/file/d/1zEjzT9FOAN9VI7nTXh5wmu-8ZvSeb72n/view?usp=sharing

 

No idea why an experienced lighting developer should receiving so much criticism.
Keep it up @aliyeredon2

 

Here are mines. Using similar test scene as yours (500 cubes with 500 dynamic lights, color temperature enabled in HD).
Build-in:


HD:


Clearly, HD pipeline is vastly superior in every department. And the lighting model is not only more complex than build-in but also better.

 

@Lu4e
No one is criticizing what aliyeredon2 did. I'm proving my point that his tests are somewhat biased. I've been working with both HD and LW pipeline sources for a couple of months now, testing and integrating a news shadow rendering systems on it. I've also worked for a very longtime with default pipeline rendering files with NGSS to know that it's old and inefficient. I've made zillions tests with simple and large scenes on all pipelines and my conclusions remains the same. Both HD and LW are superior than default pipeline. One is extremely faster (but simpler) and the other is not only faster but modern, feature rich and better.

 

230(hd) fps vs 370(built-in) fps ?

 

No,
230 buildin
370 HD.

 

Make sure :
1. Use deferred rendering path for camera (important)
2. use Low or medium shadow quality
3. disable MS AA
4. Switch to Linear

Also you can disable shadows for both of them

And finally send me your test project

 

@ aliyeredon2,
I took a look at your testing scenes they are setup differently.
To make sure everything is exactly the same for both tests, I cloned one of them and just changed scene materials and added the HD pipeline asset, here are the results:
Build-in:

HD:


Your testing scene have more lights and objects than my test scenes but the performance result is identical, around 30% - 40% faster on HD pipeline which is what I always have on all my tests cases.
And if you add shadows on top of that, default pipelines blows out immediately as it will try to create a shadowmap for every light while HD pipeline has a fixed memory budged in one atlas for all your shadowmaps, once you reach the limit it will ignore the rest without blowing out your hardware.

 

Either you have a driver bug or your hardware is too old for HD pipeline (makes use of Compute Shader heavily).

 

Send me your test scene to :
Check your settings
And test on my GPU

I'm happy to see the fact (i don't want to use any pipeline, because i don't want ti make games) . I just want to see the comparison result

 

HD is more efficient but work is still ongoing with SRP. It is illogical that builtin will be more efficient.

Matter of professional differences. It is not a criticism of the person, but his processes and findings for example would ruin my project while perhaps being good for whatever purpose he has.

 

@hippocoder
HD already outperform any render pipeline on a modern hardware. And this is just the begining.

 

@tatoforever I'm still waiting fro your test project to check settings (both pipelines)

 

Yeah, agree with that, this is expected from everyone, and Unity encourage people to test them.
No matter what results, @aliyeredon2 shouldn't be blamed, right?

Actually, I am so glad to see positive result from @tatoforever. Nice work!!
Edit:
This is not a fair test, since 500 vs 1000 is already on different parameter.
The idea of early test is to figure out possible problems for Unity to follow.
We may ignore the performance drop, but artifacts could indicate some hidden problem.
Edit2:
I added a new assumption below.

 

EXE files for both renderers

Please share your FPS + resolution + GPU model name

Use Ultra preset

https://drive.google.com/file/d/1rRMCY0_QClLNXEYi5-9_H-r6cLedRTVz/view?usp=sharing

 

No one blames him. It's just that what aliyeredon2 shows is still in early stages of development, nowhere near stable (despite him stating otherwise) and not final yet. What he compares now (performance wise), will be obsolete in the future. So, what sense does it make?

As @hippocoder said earlier:

Even Unity Tech says it in a blog post:

 

Dude, you're the only one talking about blame.

Guys - is the HD version rendering with the same feature set or are all the extra buffers being generated? I'm seeing AO there as well.

 

Should be the only one talking about cheers.
Even Unity knows how forgiving I am in past issue conversation.

Never mind, base on our past experience, reality is not always equal to expectation.

Anyway, like both of your effort here.

 

New test from these files :https://forum.unity.com/threads/hd-...ormance-comparison.526509/page-2#post-3466868
GTX 780
Built in

HD

 

Is that the project you got from @tatoforever?

*Seems the HDRP doesn't require bleeding edge hardware, as said from Joe.

edit: the anchor tags doesn't work, just search Joachim instead.

 

@tatoforever project was for mine + HD RP switch
I repeat what @tatoforever had repeated on my default project:
1. Open Built-In project
2. Create a new HD RP asset and assign into Graphics
3. Change the material to the HD Lit

The scene is simple without shadows. So the setting is not important at all (lights + simple cubes + simple default material) without skybox with default settings for both renderers + Deferred rendering path

I think maybe tatoforever result was for forward rendering path


I'm still waiting for your test results
New test results:
7580:


780:

 

You need his project to verify.

 

I said maybe
Generally, comparison in frames above 100 is not exact

 

My last test for this comparison for now :
I don't have more time for more comparisons
- Shadows : OFF
- V Sync Off
- ~ 2000 tire models
- ~1000 point lights
- Deferred rendering

Similar the first post, both are same in a heavy scene, you can only compare quality now and choose your pipeline :

 

Not sure if different beta version leads to different result, may worth a try.

 

@aliyeredon2
My tests where all done in Deferred on build-in. With Forward the performance was horrible.

 

If settings are trivial, he probably on a faulty version, could you tell him your Unity beta version?

 

2018.1b13
As a said before, we can only compare fps on lower 60 to see the actual perfoamnce
For example comparing 300 fps and 700 fps is not correct comparison.
This is my experience in optimization tests
also we must confirm editor fps using fraps. some time editor shown us 300+ fps but fraps 70-80 fps

To limit fps under 60, you can use custom resolution in game view(top left). for example 4K resolution )

 

Agree with this attitude on stress test.

Does that means relaxing the stress condition could result HDRP outperform build-in? and also artifacts free?

 

I almost overlook the fact where hippocoder sighted, as 500 vs 1000, well say this is matter of professional differences.

I believe hippocoder don't want to ruin his project with hidden problem.
So, @tatoforever could you please provide a 1000 object test and proof it is not faulty and share the project for @aliyeredon2 to audit.

@tatoforever, sorry, no offence here, just state the facts.

Edit: I added a new assumption below.

 

@Lu4e
What I meant was that what ever the amount of light or objects in my test scenes, with HD pipeline the results is always ~30% faster than Build-in pipeline.

 

Why not sharing your project out for audit?

I really wish that artifact was driver bug or outdated hardware, because both of you are excellent rendering programmer here.

 

@Lu4e,
Regarding the quick tests I did yesterday, I took one of the provided scene by Ali, clone it, change the render pipeline, change scene materials and run tests on it. One with 500 objects and 500 lights and the other one untouched and I've provided my performance results. Not sure what's going on on his hardware/OS/driver. I did my tests on i7 7700 with GTX 1070 on Win7. Everything up to date.

 

Btw,
I'm not trying to push the agenda on HD is better, just providing my performance tests with similar scenes. But if you think Build-in pipeline is better for your workflow and projects that should be enough reason to stay with.

 

Thanks much for the information, could you do a 1000 test?
Just want to confirm the robustness.

 

I truly hope I could run the test myself, sadly I don't have any working machine for the test.
Really like you guys toying around here

 

@Lu4e
This is from my test with 1000 lights from Ali provided projects.
Build-in (44fps):



HD (75fps):

 

OMG!! Almost the same fps from your previous 500 test. Unbelievable LOL
And Ali has all his hardware failed from HDRP?!

Are you on 2018.1b13 too?

 

Due to my curiosity, I did an image diff from your 500 and 1000 result, and there is totally no difference.
Hey, are you cheating us?

Edit: Dam...he shouldn't re-post the same result.

 

@aliyeredon2
Submit your case and see if Unity could reproduce this.

 

Ok, I figured out whats happening here, the previous test you have done is still 1000 test, but not untouched.

And I am still not sure if tato version is 2018.1b13.

This is almost a detective puzzle game...

 

Another test from my friend GTX 780 on 4k Res:
Built in :

HD:

 

Could you test on previous beta version?

 

1. Use 1080P or 4K res (game view->Top left)
2. Reduce light intensity on Built in pipeline to see better light effects
3. Check both camera position to be same
and capture video from your project settings : Quality window, light sources inspector and camera inspector

 

Do you have any artifact-free result from 1000 test on HDRP yet?

 

First try below exe files in your pc and send here the results if you had PC (not mac):
https://drive.google.com/file/d/1rRMCY0_QClLNXEYi5-9_H-r6cLedRTVz/view?usp=sharing

What means artifact-free ?

 

I don't have a working machine for stress testing...I am feeling so useless here.

 

From this, your HDRP result has artifact.

 

This problem happens on my 7850 (editor&build) and on my friend's GTX 780 (editor & build)
HDRP shadows and light quality is not production level now

 

But not from tato's result, thats quite weird...

I doubt you two are not on the same Unity version